Coal-fired boiler pollutant emission dynamic prediction method, device, equipment and medium

Abstract

The application discloses a coal-fired boiler pollutant emission dynamic prediction method, device, equipment and medium, and relates to the technical field of pollutant emission control. The method comprises the following steps: acquiring load parameter data and emission concentration time series data of pollutants at the outlet of a coal-fired boiler under multiple load adjustment conditions; performing time series feature extraction on the emission concentration time series data to obtain pollutant time series feature parameters corresponding to each condition; training a regression model based on a gradient boosting decision tree with the load parameter data as input and the pollutant time series feature parameters as output; inputting load parameter data of a to-be-predicted condition into the trained regression model to obtain corresponding predicted pollutant time series feature parameters, and performing time series reconstruction on the predicted pollutant time series feature parameters to obtain predicted concentration time series data of pollutants at the outlet of the boiler under the to-be-predicted condition. Through the technical scheme of the application, second-level dynamic prediction of a pollutant concentration field can be realized, and the real-time performance and accuracy of the regulation and control of a denitration system are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of pollutant emission control technology, and in particular to a method, device, equipment and medium for dynamic prediction of pollutant emissions from coal-fired boilers. Background Technology

[0002] With the increasing proportion of installed capacity from new energy sources, primarily wind and solar power, the demand for deep peak shaving in coal-fired power units is becoming increasingly urgent, leading to a significant increase in the frequency of daily load fluctuations compared to traditional operating conditions. During load changes, core parameters such as boiler burner air distribution and pulverized coal concentration field distribution exhibit minute-level fluctuations, causing significant changes in the formation characteristics of pollutants such as NOx and SOx. This strong time-varying nonlinear characteristic poses a fundamental risk of failure to traditional pollution control strategies based on steady-state operating conditions. Traditional static prediction models rely on a limited set of peak shaving test data to establish fixed mapping relationships. When the load fluctuates rapidly, they fail to capture dynamic correlations, resulting in significant prediction lags and making it difficult to meet the real-time requirements of ultra-low emission control systems. In existing technologies, prediction methods based on mechanism modeling are limited by the complexity of combustion reactions, while single machine learning models lack the ability to extract time-series features, resulting in limited accuracy in the dynamic reconstruction of pollutant concentration fields. This severely restricts the advanced control capabilities of denitrification systems and the environmental protection operation level of units. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a method, device, equipment, and medium for dynamic prediction of pollutant emissions from coal-fired boilers. This method can fundamentally solve the problems of lag and insufficient accuracy in pollutant prediction under deep peak-shaving conditions, significantly improve the real-time performance and accuracy of denitrification system control, and realize the shift from post-event control to proactive prediction. The specific solution is as follows: Firstly, this application discloses a method for dynamic prediction of pollutant emissions from coal-fired boilers, including: Acquire load parameter data and time-series data of boiler outlet pollutant emission concentrations under multiple load adjustment conditions during deep peak shaving of coal-fired boilers; The time-series data of emission concentrations are subjected to time-series feature extraction to obtain the pollutant time-series feature parameters corresponding to each load adjustment condition; Using the load parameter data as input and the pollutant time series characteristic parameters as output, a regression model based on gradient boosting decision tree is trained to establish a dynamic mapping relationship between the load parameter data and the pollutant time series characteristic parameters. The load parameter data of the operating condition to be predicted is input into the trained regression model to obtain the corresponding predicted pollutant time series characteristic parameters. The predicted pollutant time series characteristic parameters are then reconstructed to obtain the predicted concentration time series data of boiler outlet pollutants under the operating condition to be predicted.

[0004] Optionally, acquiring load parameter data and boiler outlet pollutant emission concentration time-series data under multiple load adjustment conditions during deep peak shaving of the coal-fired boiler includes: During the transition of a coal-fired boiler from initial load parameters to target load parameters, time-series data on the change of pollutant emission concentration over time are continuously collected at the boiler outlet. The time-series data collected under each load adjustment condition includes a preset number of pollutant emission concentration monitoring values ​​at equal time intervals. The pollutant emission concentration monitoring values ​​include real-time emission concentration data of one or more of nitrogen oxides, sulfur oxides, and particulate matter.

[0005] Optionally, the step of extracting time-series features from the emission concentration time-series data to obtain pollutant time-series feature parameters corresponding to each load adjustment condition includes: A convolutional neural network is used to perform convolution calculations on the time-series data of emission concentrations under each load adjustment condition, and the output after convolution calculation is used as the time-series feature parameters of the pollutants. Among them, the pollutant time-series feature parameters calculated by the convolutional neural network corresponding to a single load adjustment condition are a preset number of feature vectors.

[0006] Optionally, the process of performing convolution calculations on the emission concentration time-series data using a convolutional neural network further includes: A linear rectified activation function is introduced after each convolutional layer, and the number of channels in the convolutional layers is increased progressively.

[0007] Optionally, the load adjustment conditions are a limited number of load adjustment conditions within a preset value range; the method further includes: Among the multiple load adjustment conditions, a first preset proportion of conditions are randomly selected as training conditions, and the remaining second preset proportion of conditions are used as test conditions; the training conditions are used to train the regression model, and the test conditions are used to test and evaluate the predictive performance of the trained regression model. In the training conditions, the ratio of the load increase transition condition (adjusting from low load to high load) to the load decrease transition condition (adjusting from high load to low load) is set in a balanced manner.

[0008] Optionally, training a regression model based on a gradient boosting decision tree, using the load parameter data as input and the pollutant time-series characteristic parameters as output, includes: The initial load parameters and target load parameters under each load adjustment condition are used as inputs to the CatBoost regressor, and the pollutant time-series characteristic parameters corresponding to the load adjustment condition are used as outputs of the CatBoost regressor. The optimal hyperparameters of the CatBoost regressor are determined using a grid search method, and the CatBoost regressor is evaluated through cross-validation.

[0009] Optionally, the step of reconstructing the time-series characteristic parameters of the predicted pollutants to obtain the predicted concentration time-series data of the boiler outlet pollutants under the predicted operating condition includes: The predicted pollutant time-series feature parameters are deconvolved and restored to the same amount of sampled data as the pollutant time-series feature parameters to obtain the predicted concentration time-series data of the boiler outlet pollutants under the predicted operating condition.

[0010] Secondly, this application discloses a dynamic prediction device for pollutant emissions from a coal-fired boiler, comprising: The data acquisition module is used to acquire load parameter data and time-series data of boiler outlet pollutant emission concentrations under multiple load adjustment conditions during the deep peak shaving process of coal-fired boilers. The feature extraction module is used to extract time-series features from the emission concentration time-series data to obtain pollutant time-series feature parameters corresponding to each load adjustment condition; The prediction model building module is used to take the load parameter data as input and the pollutant time series characteristic parameters as output to train a regression model based on gradient boosting decision tree, so as to establish a dynamic mapping relationship between the load parameter data and the pollutant time series characteristic parameters. The model prediction module is used to input the load parameter data of the operating condition to be predicted into the trained regression model, obtain the corresponding predicted pollutant time series characteristic parameters, and perform time series reconstruction on the predicted pollutant time series characteristic parameters to obtain the predicted concentration time series data of boiler outlet pollutants under the operating condition to be predicted.

[0011] Thirdly, this application discloses an electronic device comprising a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the dynamic prediction method for pollutant emissions from coal-fired boilers as described above.

[0012] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the dynamic prediction method for pollutant emissions from coal-fired boilers as described above.

[0013] The beneficial effects of this application are as follows: By specifically extracting time-series features from pollutant emission concentration time-series data, the dynamic evolution law of pollutant concentration with changes in combustion conditions during rapid load fluctuations is effectively captured. Based on this, the powerful nonlinear fitting capability of the gradient boosting decision tree model is utilized to establish an accurate mapping relationship from load parameter data to pollutant time-series feature parameters, avoiding the significant prediction lag problem of traditional static prediction models that rely on a fixed mapping relationship under steady-state conditions when the load changes rapidly. The organic integration of these two approaches overcomes the inherent defects of single-mechanism models in describing the strong time-varying nonlinear characteristics of deep peak-shaving processes and the insufficient ability of single machine learning models to extract time-series features, further realizing accurate prediction of pollutant emission concentrations from coal-fired boilers across the entire load range. In actual prediction, only the load parameter data of the operating condition to be predicted needs to be directly input into the model, and the predicted value of pollutant emission concentration for the future period can be obtained immediately after the load command is issued, providing a forward-looking control basis for the Selective Catalytic Reduction (SCR) denitrification system. Compared to traditional methods that only respond passively after pollutant concentrations actually change, this invention achieves a shift from "post-event control" to "proactive prediction," significantly improving the timeliness and accuracy of denitrification system control and effectively reducing the risk of exceeding emission standards due to control lag. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a flowchart of a method for dynamic prediction of pollutant emissions from a coal-fired boiler disclosed in this application; Figure 2 This is a schematic diagram of a CNN-CatBoost hybrid prediction architecture disclosed in this application; Figure 3 This is a schematic diagram of the structure of a dynamic prediction device for pollutant emissions from a coal-fired boiler disclosed in this application. Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Currently, methods for predicting the concentration of pollutants (such as nitrogen oxides) in flue gas from coal-fired power units have two major shortcomings, resulting in the inability to accurately and proactively regulate the denitrification system and affecting the environmental performance of the units. Firstly, prediction methods based on mechanistic modeling establish physical models based on chemical reaction principles. However, due to the extreme complexity of combustion reactions within the furnace (high temperature, turbulence, multiphase flow), these models are difficult to describe precisely with mathematical formulas, leading to inaccurate predictions. Secondly, relying solely on a machine learning algorithm (such as a backpropagation neural network) to process data that changes over time cannot fully capture the dynamic changes in pollutant concentrations (such as fluctuations, delays, and correlations), resulting in insufficient prediction accuracy. Therefore, these two reasons make it difficult to accurately predict the concentration of pollutants in boiler flue gas, thus limiting the proactive regulation of the denitrification system and the environmental compliance capabilities of the units.

[0018] To this end, this application provides a dynamic prediction scheme for pollutant emissions from coal-fired boilers, which can establish a nonlinear dynamic correlation between load and pollutant parameters, effectively overcoming the excessive reliance of traditional models on steady-state operating data; it can predict the pollutant generation trend based on the advance information of the target load and the current original load after the load command is issued, avoiding the control lag caused by the passive response of traditional methods after the actual change of pollutant concentration, and effectively reducing the risk of excessive emissions caused by control delay.

[0019] This invention discloses a method for dynamic prediction of pollutant emissions from coal-fired boilers. (See also...) Figure 1 As shown, the method includes: Step S11: Obtain load parameter data and time series data of boiler outlet pollutant emission concentration under multiple load adjustment conditions during the deep peak shaving process of the coal-fired boiler.

[0020] In this embodiment, the load adjustment condition refers to the process by which a coal-fired boiler transitions from a relatively stable initial load to another target load. This includes a load increase transition condition (adjusting from low load to high load) and a load decrease transition condition (adjusting from high load to low load). The initial load parameters and target load parameters are preferably expressed as percentages relative to the rated load, both falling within the 30% to 100% load range. That is, the highest load for all conditions is 100% load, and the lowest load is 30% load.

[0021] When on-site testing conditions are available, deep peak-shaving tests can be conducted according to a pre-designed load adjustment plan, and load parameter data and boiler outlet pollutant emission concentration time-series data, as well as corresponding load commands, can be recorded simultaneously under multiple load adjustment conditions. When on-site testing conditions are not available, data can be acquired in real time through a distributed control system (DCS).

[0022] In one specific implementation scheme, the data acquisition process described above is performed as follows: During the transition of the coal-fired boiler from initial load parameters to target load parameters, time-series data on the change of pollutant emission concentration over time are continuously collected at the boiler outlet. It is evident that for each load adjustment condition, at least the initial load parameters, target load parameters, and time-series data on the change of pollutant emission concentration over time are recorded; if necessary, the start and end times of the condition can also be recorded to align the load change process with the pollutant response process. The time-series data collected under each load adjustment condition includes a preset number of pollutant emission concentration monitoring values ​​at equal time intervals. For example, for each load adjustment condition, 1000 to 2000 monitoring values ​​at equal time intervals are continuously obtained at the boiler outlet. The pollutant emission concentration monitoring values ​​include real-time emission concentration data for one or more of nitrogen oxides, sulfur oxides, and particulate matter.

[0023] It should be noted that the sampling interval can follow the inherent sampling cycle of the on-site continuous emission monitoring system or DCS. If the number of effective sampling points for each operating condition is not completely consistent, a uniform target sampling length can be selected within the aforementioned range of 1000-2000 points, and the sample length can be made consistent by truncating the effective time window or resampling at equal intervals. This processing is only for facilitating batch input into the network and does not change the basic trend of pollutant concentration changes over time.

[0024] In this embodiment, the load adjustment conditions are a limited number of load adjustment conditions within a preset value range. In one specific implementation, the total number of load adjustment conditions is set to 80 to 200 groups. To enable the model to be applicable to both load increase and load decrease scenarios, the ratio of load increase transition conditions to load decrease transition conditions is balanced in the training conditions. That is, 50% of the conditions involve adjusting from a relatively low load to a relatively high load, and 50% involve adjusting from a relatively high load to a relatively low load.

[0025] Among multiple load adjustment scenarios, a first preset proportion (e.g., 80%) is randomly selected as the training scenario, and the remaining second preset proportion (i.e., 20%) is used as the test scenario. The training scenario is used to train the regression model, and the test scenario is used to test and evaluate the predictive performance of the trained regression model. Preferably, a random partitioning method based on the load adjustment direction can be adopted, so that both the training set and the test set include load increase transition scenarios and load decrease transition scenarios.

[0026] Step S12: Extract time-series features from the emission concentration time-series data to obtain pollutant time-series feature parameters corresponding to each load adjustment condition.

[0027] In this embodiment, a convolutional neural network is used to perform convolution calculations on the time-series emission concentration data under various load adjustment conditions, and the output after convolution calculation is used as the pollutant time-series feature parameters. The purpose of this step is to utilize the feature extraction capability of the convolutional neural network for one-dimensional time-series signals to compress and transform the acquired raw pollutant emission concentration time-series data into a set of low-dimensional numerical feature parameters that can characterize the core dynamic pattern of the time-series data.

[0028] It should be noted that a Rectified Linear Activation Function (ReLU) is introduced after each convolutional layer to enhance the network's ability to express nonlinear temporal changes; and the number of channels in the convolutional layers is increased progressively with increasing network depth to capture multi-scale temporal patterns. This allows shallow convolutions to focus on capturing local changes at short timescales, while deeper convolutions focus on combining and forming trend features at longer timescales. Specifically, the pollutant temporal feature parameters calculated by the convolutional neural network for a single load adjustment condition are a preset number of feature vectors. In one specific implementation, the pollutant temporal feature parameters calculated by the convolutional neural network for a single load adjustment condition are 10-40 sets.

[0029] Step S13: Using load parameter data as input and pollutant time series characteristic parameters as output, train a regression model based on gradient boosting decision tree to establish a dynamic mapping relationship between load parameter data and pollutant time series characteristic parameters.

[0030] In this step, the initial load parameters and target load parameters for each training condition are used as inputs, and the pollutant time-series feature parameters obtained in step S12 are used as supervised outputs to train a regression model based on Gradient Boosting Decision Tree (GBDT). Preferably, the regression model uses a CatBoost regressor. CatBoost can establish a nonlinear mapping relationship between the two load parameters and high-dimensional time-series feature parameters by constructing a decision tree round by round and fitting the residuals.

[0031] In a feasible implementation, the regression model can also employ XGBoost or LightGBM. Both models possess strong nonlinear fitting capabilities and the ability to handle high-dimensional data, offering different computational efficiencies and memory footprints while maintaining prediction accuracy. However, it should be noted that CatBoost has unique advantages in handling categorical features and preventing overfitting; therefore, the hyperparameter optimization strategy needs to be adjusted accordingly when replacing it.

[0032] During training, a grid search method is used to determine the optimal hyperparameters of the CatBoost regressor. Parameters that can be included in the search include tree depth, learning rate, number of iterations, and regularization-related parameters; the specific candidate range can be determined based on the number of samples and feature dimensions. Five-fold cross-validation is performed on each candidate hyperparameter combination to evaluate the CatBoost regressor. Specifically, the training conditions are divided into five subsets, with one subset used as the validation set and the remaining four subsets used as the fitting data in turn. The best-performing hyperparameter combination is selected by combining the results of the five validation runs.

[0033] After model training is complete, 20% of the pre-reserved test conditions are input into the trained CatBoost regressor to obtain the predicted pollutant time-series feature parameters for the test conditions, which are then deconvolved to obtain the predicted concentration time-series data. The model is evaluated by comparing the predicted concentration time-series data with the corresponding measured concentration time-series data. Under the data configuration of this embodiment, the determination coefficient R² of the test results is greater than 0.95, significantly outperforming conventional machine learning methods such as random forests and support vector machines.

[0034] Step S14: Input the load parameter data of the operating condition to be predicted into the trained regression model to obtain the corresponding predicted pollutant time series characteristic parameters, and perform time series reconstruction on the predicted pollutant time series characteristic parameters to obtain the predicted concentration time series data of boiler outlet pollutants under the operating condition to be predicted.

[0035] In this embodiment, the initial load parameters and target load parameters of the operating condition to be predicted are received. The operating condition to be predicted can be a load combination that did not appear in the training conditions, as long as the initial load and target load are within the 30%~100% load range. The load parameter data of the operating condition to be predicted is input into the trained CatBoost regressor to obtain the predicted pollutant time-series characteristic parameters corresponding to the operating condition to be predicted.

[0036] Subsequently, the predicted pollutant time-series feature parameters are reconstructed, i.e., deconvolution is performed on the predicted pollutant time-series feature parameters. These parameters are then input into the deconvolution reconstruction network corresponding to the convolutional feature extraction network in step S12. During deconvolution, the temporal resolution and channel information are restored step-by-step, ultimately restoring the sampled data to the same amount as the original pollutant time-series feature parameters for this type of operating condition. Specifically, this results in 1000-2000 predicted concentration values ​​at equal time intervals, yielding the predicted concentration time-series data of the boiler outlet pollutants under the predicted operating condition. If multiple pollutants are predicted simultaneously, the output sequences are the predicted concentrations of nitrogen oxides, sulfur oxides, and particulate matter, respectively.

[0037] Because the model input includes the current initial load parameters and the target load parameters, it can predict the subsequent pollutant emission trend immediately after the load command is issued, without waiting for a significant change in the pollutant concentration at the boiler outlet. The prediction results can serve as feedforward information for the SCR denitrification system, allowing for advance adjustment of ammonia injection or other pollutant control parameters, thereby reducing the risk of short-term excessive emissions caused by control lag.

[0038] The beneficial effects of this application are as follows: By specifically extracting time-series features from pollutant emission concentration time-series data, the dynamic evolution law of pollutant concentration with changes in combustion conditions during rapid load fluctuations is effectively captured. Based on this, the powerful nonlinear fitting capability of the gradient boosting decision tree model is utilized to establish an accurate mapping relationship from load parameter data to pollutant time-series feature parameters, avoiding the significant prediction lag problem of traditional static prediction models that rely on a fixed mapping relationship under steady-state conditions when the load changes rapidly. The organic integration of these two approaches overcomes the inherent defects of single-mechanism models in describing the strong time-varying nonlinear characteristics of deep peak-shaving processes and the insufficient ability of single machine learning models to extract time-series features, further realizing accurate prediction of pollutant emission concentrations from coal-fired boilers across the entire load range. In actual prediction, only the load parameter data of the operating condition to be predicted needs to be directly input into the model, and the predicted value of pollutant emission concentration for the future period can be obtained immediately after the load command is issued, providing a forward-looking control basis for the Selective Catalytic Reduction (SCR) denitrification system. Compared to traditional methods that only respond passively after pollutant concentrations actually change, this invention achieves a shift from "post-event control" to "proactive prediction," significantly improving the timeliness and accuracy of denitrification system control and effectively reducing the risk of exceeding emission standards due to control lag.

[0039] like Figure 2The diagram illustrates a CNN-CatBoost hybrid prediction architecture based on the above embodiment, used for dynamically predicting pollutant emissions from coal-fired boilers. The architecture uses multiple known load adjustment scenarios as training samples, each scenario being characterized by initial load parameters, target load parameters, and corresponding time-series data of pollutant emission concentrations at the boiler outlet. First, a convolutional neural network compresses the high-dimensional, long-term time-series data into time-series feature parameters that characterize the dynamic changes in pollutants. Then, a CatBoost regressor learns the mapping relationship between the load parameters and these time-series feature parameters. In actual prediction, only the initial and target load parameters of the scenario to be predicted need to be input to predict the corresponding time-series feature parameters, which are then reconstructed into pollutant concentration time-series data through deconvolution. The pollutants can be one of nitrogen oxides, sulfur oxides, and particulate matter, or two or three of them simultaneously. When predicting multiple pollutants simultaneously, the concentration sequences of different pollutants can be input into the convolutional neural network as different input channels, or corresponding convolutional feature extraction networks and regression models can be established separately.

[0040] As can be seen, the CNN-CatBoost hybrid architecture effectively captures the temporal dynamic characteristics of pollutant concentration fluctuations with load under deep peak shaving conditions by using a convolutional neural network (CNN) to extract multi-scale features from pollutant time-series data. Then, a CatBoost regressor establishes a nonlinear dynamic mapping relationship between load parameters and feature parameters. Compared to traditional static prediction models that rely on a limited set of peak-shaving test data to establish a fixed mapping relationship and suffer from significant prediction lag due to the inability to capture dynamic correlations during rapid load fluctuations, this invention can achieve second-level dynamic prediction of pollutant concentration fields, with an R² value exceeding 0.95, significantly outperforming conventional machine learning methods such as random forests and support vector machines. Existing prediction methods based on mechanism modeling are limited by the complexity of combustion reactions, making it difficult to accurately describe the impact of minute-level fluctuations in core parameters such as boiler burner air distribution and pulverized coal concentration field distribution on pollutant generation characteristics during deep peak shaving. Furthermore, single machine learning models lack sufficient ability to extract temporal features, resulting in limited accuracy in the dynamic reconstruction of pollutant concentration fields. This invention effectively overcomes the inherent defects of the two methods mentioned above by organically combining the temporal feature extraction capability of CNN with the strong nonlinear fitting capability of CatBoost, and achieves accurate prediction of emission concentrations of pollutants such as NOx and SOx in the full load range of 30%-100%.

[0041] In one feasible implementation, a mechanism-data hybrid driven model can be used to replace the CNN-CatBoost hybrid prediction architecture. Specifically, based on a boiler combustion mechanism model (such as a boiler combustion mechanism model, an in-furnace NOx generation and reduction mechanism model, etc.), machine learning algorithms are used to compensate for and correct the unmodeled dynamics and biases of the mechanism model. This approach combines the interpretability of mechanism models with the fitting accuracy of data-driven models; however, the construction of mechanism models requires a deep understanding of the physicochemical processes of boiler combustion, resulting in a long development cycle.

[0042] In summary, this invention can directly acquire load parameters and pollutant concentration time-series data in real time through a distributed control system (DCS) even without on-site experimental conditions. Model training can be completed with only 80 to 200 sets of sample data under limited operating conditions, significantly reducing the cost and data acquisition difficulty of on-site experiments under deep peak-shaving conditions. Simultaneously, the training conditions evenly cover both load increase and decrease transition processes, ensuring the model has good generalization ability under bidirectional load variation scenarios. By improving the prediction accuracy and response speed of pollutant emission concentrations, this invention can effectively guide the precise ammonia injection of denitrification systems, reducing the excessive addition of reducing agents (such as ammonia water or urea), and lowering the operating costs of denitrification systems while ensuring ultra-low emission standards are met, thus achieving the dual goals of environmental compliance and economic operation.

[0043] Accordingly, this application also discloses a dynamic prediction device for pollutant emissions from coal-fired boilers, see [link to relevant documentation]. Figure 3 As shown, the device includes: Data acquisition module 11 is used to acquire load parameter data and time series data of boiler outlet pollutant emission concentration under multiple load adjustment conditions during deep peak shaving of coal-fired boilers. The feature extraction module 12 is used to extract time-series features from the emission concentration time-series data to obtain pollutant time-series feature parameters corresponding to each load adjustment condition; The prediction model building module 13 is used to take the load parameter data as input and the pollutant time series characteristic parameters as output to train a regression model based on gradient boosting decision tree, so as to establish a dynamic mapping relationship between the load parameter data and the pollutant time series characteristic parameters. The model prediction module 14 is used to input the load parameter data of the operating condition to be predicted into the trained regression model, obtain the corresponding predicted pollutant time series characteristic parameters, and perform time series reconstruction on the predicted pollutant time series characteristic parameters to obtain the predicted concentration time series data of the boiler outlet pollutants under the operating condition to be predicted.

[0044] For more detailed information on the working process of each of the above modules, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.

[0045] Therefore, the above-described scheme in this embodiment effectively captures the dynamic evolution of pollutant concentrations under varying combustion conditions during rapid load fluctuations by extracting specific time-series features from pollutant emission concentration time-series data. Based on this, the powerful nonlinear fitting capability of the gradient boosting decision tree model is utilized to establish a precise mapping relationship from load parameter data to pollutant time-series feature parameters. This avoids the significant prediction lag that traditional static prediction models, which rely on a fixed mapping relationship under steady-state conditions, exhibit when the load changes rapidly. The organic integration of these two approaches overcomes the inherent limitations of single-mechanism models in describing the strong time-varying nonlinear characteristics of deep peak-shaving processes and the insufficient time-series feature extraction capability of single machine learning models, further achieving accurate prediction of pollutant emission concentrations from coal-fired boilers across the entire load range. In actual prediction, only the load parameter data of the operating condition to be predicted needs to be directly input into the model to obtain the predicted pollutant emission concentration for the future period immediately after the load command is issued, providing a forward-looking control basis for the Selective Catalytic Reduction (SCR) denitrification system. Compared to traditional methods that only respond passively after pollutant concentrations actually change, this invention achieves a shift from "post-event control" to "proactive prediction," significantly improving the timeliness and accuracy of denitrification system control and effectively reducing the risk of exceeding emission standards due to control lag.

[0046] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0047] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the dynamic prediction method for pollutant emissions from coal-fired boilers disclosed in any of the foregoing embodiments.

[0048] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0049] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored on it can include an operating system 221, computer programs 222, and data 223, etc. The data 223 can include various types of data. The storage method can be temporary storage or permanent storage.

[0050] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the dynamic prediction method for pollutant emissions from a coal-fired boiler executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0051] Furthermore, this application also discloses a computer-readable storage medium, which includes random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, magnetic disks, optical disks, or any other form of storage medium known in the art. The computer program, when executed by a processor, implements the aforementioned dynamic prediction method for pollutant emissions from coal-fired boilers. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0052] Furthermore, embodiments of this application also provide a computer program product, including a computer program / instructions, which, when executed by a processor, implements any of the above-described methods for dynamic prediction of pollutant emissions from coal-fired boilers.

[0053] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0054] The steps of the dynamic prediction method or algorithm for pollutant emissions from coal-fired boilers described in conjunction with the embodiments disclosed herein can be implemented directly using hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0055] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0056] The above provides a detailed description of the method, apparatus, equipment, and medium for dynamic prediction of pollutant emissions from coal-fired boilers provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for dynamic prediction of pollutant emissions from coal-fired boilers, characterized in that, include: Acquire load parameter data and time-series data of boiler outlet pollutant emission concentrations under multiple load adjustment conditions during deep peak shaving of coal-fired boilers; The time-series data of emission concentrations are subjected to time-series feature extraction to obtain the pollutant time-series feature parameters corresponding to each load adjustment condition; Using the load parameter data as input and the pollutant time series characteristic parameters as output, a regression model based on gradient boosting decision tree is trained to establish a dynamic mapping relationship between the load parameter data and the pollutant time series characteristic parameters. The load parameter data of the operating condition to be predicted is input into the trained regression model to obtain the corresponding predicted pollutant time series characteristic parameters. The predicted pollutant time series characteristic parameters are then reconstructed to obtain the predicted concentration time series data of the boiler outlet pollutants under the operating condition to be predicted.

2. The method for dynamic prediction of pollutant emissions from coal-fired boilers according to claim 1, characterized in that, The acquisition of load parameter data and boiler outlet pollutant emission concentration time-series data under multiple load adjustment conditions during deep peak shaving of coal-fired boilers includes: During the transition of a coal-fired boiler from initial load parameters to target load parameters, time-series data on the change of pollutant emission concentration over time are continuously collected at the boiler outlet. The time-series data collected under each load adjustment condition includes a preset number of pollutant emission concentration monitoring values ​​at equal time intervals. The pollutant emission concentration monitoring values ​​include real-time emission concentration data of one or more of nitrogen oxides, sulfur oxides, and particulate matter.

3. The method for dynamic prediction of pollutant emissions from coal-fired boilers according to claim 1, characterized in that, The step of extracting time-series features from the emission concentration time-series data to obtain pollutant time-series feature parameters corresponding to each load adjustment condition includes: A convolutional neural network is used to perform convolution calculations on the time-series data of emission concentrations under each load adjustment condition, and the output after convolution calculation is used as the time-series feature parameters of the pollutants. Among them, the pollutant time-series feature parameters calculated by the convolutional neural network corresponding to a single load adjustment condition are a preset number of feature vectors.

4. The method for dynamic prediction of pollutant emissions from coal-fired boilers according to claim 3, characterized in that, The process of using a convolutional neural network to perform convolution calculations on the time-series emission concentration data also includes: A linear rectified activation function is introduced after each convolutional layer, and the number of channels in the convolutional layers is increased progressively.

5. The method for dynamic prediction of pollutant emissions from coal-fired boilers according to claim 1, characterized in that, The load adjustment conditions are a limited number of load adjustment conditions within a preset value range; the method further includes: Among the multiple load adjustment conditions, a first preset proportion of conditions are randomly selected as training conditions, and the remaining second preset proportion of conditions are used as test conditions; the training conditions are used to train the regression model, and the test conditions are used to test and evaluate the predictive performance of the trained regression model. In the training conditions, the ratio of the load increase transition condition (adjusting from low load to high load) to the load decrease transition condition (adjusting from high load to low load) is set in a balanced manner.

6. The method for dynamic prediction of pollutant emissions from coal-fired boilers according to claim 2, characterized in that, The step of training a regression model based on a gradient boosting decision tree, using the load parameter data as input and the pollutant time-series characteristic parameters as output, includes: The initial load parameters and target load parameters under each load adjustment condition are used as inputs to the CatBoost regressor, and the pollutant time-series characteristic parameters corresponding to the load adjustment condition are used as outputs of the CatBoost regressor. The optimal hyperparameters of the CatBoost regressor are determined using a grid search method, and the CatBoost regressor is evaluated through cross-validation.

7. The method for dynamic prediction of pollutant emissions from coal-fired boilers according to any one of claims 1 to 6, characterized in that, The step of reconstructing the time-series characteristic parameters of the predicted pollutants to obtain the predicted concentration time-series data of the boiler outlet pollutants under the predicted operating condition includes: The predicted pollutant time-series feature parameters are deconvolved and restored to the same amount of sampled data as the pollutant time-series feature parameters to obtain the predicted concentration time-series data of the boiler outlet pollutants under the predicted operating condition.

8. A dynamic prediction device for pollutant emissions from a coal-fired boiler, characterized in that, include: The data acquisition module is used to acquire load parameter data and time-series data of boiler outlet pollutant emission concentrations under multiple load adjustment conditions during the deep peak shaving process of coal-fired boilers. The feature extraction module is used to extract time-series features from the emission concentration time-series data to obtain pollutant time-series feature parameters corresponding to each load adjustment condition; The prediction model building module is used to take the load parameter data as input and the pollutant time series characteristic parameters as output to train a regression model based on gradient boosting decision tree, so as to establish a dynamic mapping relationship between the load parameter data and the pollutant time series characteristic parameters. The model prediction module is used to input the load parameter data of the operating condition to be predicted into the trained regression model, obtain the corresponding predicted pollutant time series characteristic parameters, and perform time series reconstruction on the predicted pollutant time series characteristic parameters to obtain the predicted concentration time series data of boiler outlet pollutants under the operating condition to be predicted.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the dynamic prediction method for pollutant emissions from coal-fired boilers as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein the computer programs, when executed by a processor, implement the dynamic prediction method for pollutant emissions from coal-fired boilers as described in any one of claims 1 to 7.

Images